The problem that I encountered is that the program that I have written times out for the new 2.0 beta. I upgraded the driver today along with the card. My system use to have a 8500GT and a 8800Ultra, however it now has a 8500GT and a GTX 280. The 8500 is used for display and no display is extended for use of the 280.
The problem is, that before the upgrade the watchdog did not stop the secondary card. This is not true anymore. The question that I have is: Can I disable the watchdog or is there something that I have missed after the install of the new card? And why is watchdog more involved for the 2.0 beta?
The 280 is not the default and it is not used as a display.
Two cards, second card is used for CUDA experiments and windows display is NOT extended onto it, it is used for gpu calculations only. Under CUDA 1.1 kernels are not timed out for the second card, under CUDA 2.0 Beta2 they are.
This is significant problem, watchdog must be off for the second card or it must be switchable.
NVIDIA guys: looks like this is a serious problem of CUDA 2.0.
Well, here is a update. I have not been able to get it to work so I have went back to the 8800 with the 1.1 driver. I do beleive it is a driver problem. I got a blue screen of death when I booted with the 2.0 driver and the 8800, so I had to go to safe mode and go back to the 1.1. The watchdog is the problem from the driver of 2.0. And the problem is that only driver 2.0 beta supports the gtx 280. So until I see that that problem is fixed for the 2.0 driver, I will have to continue with the 8800 and 1.1 driver.
Prelution
CUDA developers, please shin some light on this problem.